sstewart@NBER.ORG wrote: >I have run the same probit using PROC PROBIT and PROC LOGISTIC link=normit, >and I get identical coeficients, but the values for intercepts 2, 3, and 4 >are different. Does anyone know why this is the case? We particularly care >about the most accurate value of the 4th intercept (the value among people >with zero on all of the binary predcitor variables). > >Intercepts from PROC Probit: > >Intercept 1 -0.2966 0.0001 -0.2967 -0.2964 1.171E7 <.0001 >Intercept2 1 0.9721 0.0001 0.9719 0.9723 1.236E8 <.0001 >Intercept3 1 2.0368 0.0001 2.0365 2.0370 2.138E8 <.0001 >Intercept4 1 2.9892 0.0002 2.9887 2.9897 1.564E8 <.0001 > >Intercepts from PROC Logistic link=normit: > >Intercept 1 1 -0.2966 0.000087 11714433.3 <.0001 >Intercept 2 1 0.6755 0.000091 55221920.9 <.0001 >Intercept 3 1 1.7402 0.000135 165280320 <.0001 >Intercept 4 1 2.6926 0.000238 128041714 <.0001 > >thank you, >Susan > >The SAS progam and full results are below: > >proc probit; class rthlth42; >where rthlth42 ne -8; model rthlth42 = work rec workrec adl >walk bend walkbend dep anx depanx seed heard; output out=mpred >p=yhat; weight SQPQW00F; > >proc logistic outest=logres; class rthlth42; >where rthlth42 ne -8; model rthlth42 = work rec workrec adl >walk bend walkbend dep anx depanx seed heard /link=normit technique=newton; >weight SQPQW00F;

I see that Shiling has already pointed out that the two procs parametrize the model differently, giving you different 'intercept' values. And he has pointed out one way to handle that problem.

But I wanted to point out two other things:

[1] You are using weights. Are these sample weights from a survey sample? If so, then you should not be using EITHER of these procs. You should be using PROC SURVEYLOGISTIC. And you should be incorporating *all* the survey design effects of the study in order to get the right results.

[2] The difference in parameterization is easy to see. Your results for intercepts 2 to 4 in PROC LOGISTIC differ from those in PROC PROBIT by *exactly* the size of intercept 1.

HTH, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ On the road to retirement? Check out MSN Life Events for advice on how to get there! http://lifeevents.msn.com/category.aspx?cid=Retirement